More than 750,000 office and administrative support jobs are projected to vanish from the US economy by 2035. Silicon Valley Girl built a ranked countdown from Bureau of Labor Statistics projections, cross-referencing each occupation against AI exposure scores drawn from five research sources, including Microsoft Copilot usage data mapped to real workplace tasks, to find which careers are actually growing and what they pay. The answer surprised even her: of the 206 occupations with the highest AI exposure, 141 are still projected to grow over the next decade. That is 68%. High AI exposure and high replacement risk are not the same thing.
The countdown from number ten to number one
At number ten sit management analysts, the consultants companies hire to figure out how to do work better. They earn just over $100,000 a year and are projected to add 109,000 jobs, with 94,000 openings every year. The reason is blunt: every company is trying to restructure itself around AI and most have no idea how. The restructuring is the product.
Number nine is the entry that almost no one puts on these lists: health sciences professors and clinical educators, the people who train nurses, pharmacists, physical therapists, and physician assistants. Healthcare is projected to account for roughly a third of all new American jobs this decade, and every one of those workers has to be taught by a credentialed human. Clinical training legally cannot be scaled with a chatbot; supervised human hours are required by law. The barrier is real: a doctoral or professional degree is needed, making this a decade-long play rather than a quick career pivot.
Number eight is construction managers, adding 55,000 jobs with nearly 50,000 openings a year. The AI connection is physical. Data centers, chip fabrication plants, and AI factories all begin as construction projects. Jensen Huang said on Fox Business: ‘Don’t forget chip plants are being created, packaging, computer plants, all of the AI factories being created. Hundreds of thousands of jobs are being created as we speak.’
Number seven is data scientists, growing 34.6%, from 275,000 to an estimated 371,000 by 2035. AI automated the middle of the job: the cleaning, the boilerplate, the first-pass model. It did not touch the two ends: deciding which question is worth asking, and deciding whether the answer is trustworthy enough to act on.
Number six is medical and health services managers, adding 155,000 jobs at a 24.2% growth rate, with 62,000 openings a year. A bachelor’s degree suffices; a medical degree does not. Healthcare is exploding from aging demand while simultaneously being the most fragmented, most regulated, most administratively broken industry in the developed world, and it is being digitized badly and expensively right now.
Number five is information security analysts, up 21%. AI made attacking cheap: phishing in fluent English at scale, voice cloning, automated vulnerability scanning. Companies are pushing AI into everything, creating categories of risk no one has a playbook for yet. At roughly 193,000 people total, this is a smaller occupation than its reputation implies, with 14,000 openings a year, and genuinely competitive, but it offers the most accessible high-paying path without a heavy degree requirement.
Number four is nurse practitioners at $132,000 a year, 29,000 openings annually, and plus 41%: the single fastest-growing occupation in the United States. A physician shortage is pushing more primary care toward nurses with master’s degrees. AI carries high exposure here and the occupation is still the fastest-growing job in the country. As Silicon Valley Girl put it: ‘AI can draft notes, flag interactions, summarize a chart. It cannot examine a patient, hold legal responsibility for a diagnosis, or sit with a frightened person for 20 minutes.’
Number three is physician assistants: $136,000 a year, 21% growth, 35,000 new jobs, and the lowest AI exposure quartile out of 831 occupations scored. That combination, high pay, strong growth, and minimal AI replacement risk, appears almost nowhere else in the dataset.
Number two is software developers at $135,000 a year, 10% growth, 95,000 openings a year, and the largest projected absolute gain of any occupation paying over $100,000: 174,000 new jobs, taking total headcount from roughly 1.7 million to roughly 1.9 million by 2035. The picture is mixed underneath that headline. TechCrunch, citing Signal Fire data, found big tech hiring down about 25% from 2019, but engineering hiring down only 11%, and engineers rising from 46% to 55% of new hires. Indeed’s hiring lab found US software postings up 15% since Claude Code launched while overall postings fell 7%, but 71% of that gain came from senior roles. The job is growing; the entry point is narrowing.
Number one is computer and information systems managers at $175,000 a year, plus 15.8%, 108,000 new jobs. IT directors, engineering managers, and CTOs of midsize companies. The work is rated very high AI exposure and it is still growing near the top of the pay scale because the scarce resource in 2026 is not the ability to produce technical output. AI made that cheap. The scarce resource is judgment under uncertainty and accountability for the outcome.
The five occupations shrinking fastest
All five carry very high AI exposure ratings. Customer service representatives are down 5% in percentage terms, but that translates to 141,000 fewer jobs from a base of 2.6 million. Goldman Sachs research found US call center employment running 39% below trend. Even while shrinking, this occupation still posts 289,000 openings a year: declining does not mean disappearing, it means the ceiling is falling.
Bookkeeping, accounting, and auditing clerks are down 5%, but on a base of 1.5 million people that means 85,000 jobs gone. Payroll and timekeeping clerks are down 15%, losing 25,000 of 159,000 positions at $58,000 a year above the national median. Data entry keyers are down 25% at $41,000 a year. And the occupation with the single fastest projected percentage decline in the US economy is word processors and typists, down 34%.
Stanford economist Eric Bernovson, who studies what AI actually does to wages and hiring, offered the cleanest framework for identifying where any job is headed: nearly every project divides into three parts, defining the question, executing once the question is clear, and evaluating whether the output did what was needed. AI agents are getting very good at the middle step. The parts that stay human are the two ends.
The 28% gap worth sitting with
Right now, 28% of Americans use AI at work at all. In Singapore, it is 61%. In the UAE, it is 64%. When researchers ranked what people actually do with AI, running an agent did not make the top ten, not because it is technically hard, but because almost no one has sat down and tried it once.
The 174,000 projected new software jobs sit alongside a narrowing entry point for junior engineers, and nobody has fully mapped what the bottom of that funnel looks like by 2030.
